Contributor
Posts: 28

# Odds ratio - univariate and logistic regression points in different ways

Dear anyone

I am using SAS 9.4, enterprise guid 6.1. I am looking at the risk of taking medicine X if you have symptom A, B and C. I have to groups: I Use of medicine X and II no use of medicine X. All symptoms are more frequent in the group that uses medicine.  When I do univariate analysis I get the following odds ratio for taking medicine X: 3.4, 2.3, and 1.2. When I do a logistic regression I get 0.3, 0.6 and 1.2

I can appreciate that I get something close to the inverse OR depending of my coding. But how can symptom C point in another direction than the rest. To me it does not make any sense. Can anyone please help me.

Se tables for details

Sincerely

Anders

 Univariate Symp A Total 0 1 med 0 Frequency 513 144 657 Row Pct 78.08 21.92 1 Frequency 287 270 557 Row Pct 51.53 48.47 Total Frequency 800 414 1214 OR=3.4 SympB Total 0 1 med 0 Frequency 548 109 657 Row Pct 83.41 16.59 1 Frequency 382 175 557 Row Pct 68.58 31.42 Total Frequency 930 284 1214 OR=2.3 SympC Total 0 1 med 0 Frequency 598 59 657 Row Pct 91.02 8.98 1 Frequency 497 60 557 Row Pct 89.23 10.77 Total Frequency 1095 119 1214 OR=1.2

 Logistic regression Odds Ratio Estimates and Wald Confidence Intervals Effect Unit Estimate 95% Confidence Limits SympA 0 vs 1 1 0.334 0.257 0.435 SympB 0 vs 1 1 0.627 0.468 0.842 SympC 0 vs 1 1 1.157 0.771 1.734
Super User
Posts: 13,542

## Re: Odds ratio - univariate and logistic regression points in different ways

Code used to generate the summary will be helpfule as there are enough options involved that just the output is not enough for diagnostics.

Contributor
Posts: 28

## Re: Odds ratio - univariate and logistic regression points in different ways

Dear ballardw

I have talked to a statistician and the 'problem' was just confounding.

Sincerely Anders

Super Contributor
Posts: 339

## Re: Odds ratio - univariate and logistic regression points in different ways

My first haunch would be that you did not use the DESCENDING option in your proc logistic so it's modeling the odds ratio of going from state "taking the medicine" to "not taking the medicine"

Contributor
Posts: 28

Dear Vince